Compact Letter Display from a matrix of significancies or by hand

Question

I am running a multiple pairwise comparison in R. I'm using the survival package survminer. I'm using the function: pairwise_survdiff {survminer}

It gives the pairwise comparisons with significance as expected, but doesn't seem to have a way to give a compact letter display (CLD) of the results. I'm looking at pairs of 19 levels. I ended up printing the results, putting them into excel by hand and then doing letters by hand. But now I need to do it again and am hoping for an easier way.

Can I have R do a CLD from the pairwise_survdiff {survminer} results directly?

Baring that

Is there a way to get it to print results into a table that can be read by a spreadsheet?
If I make the logic matrix by hand, how do I have R take that and turn it into a CLD?

And 4) If I'm doing it all by hand, I'm wondering if there is a more compact method of showing this list of comparisons. Can I eliminate any of these letters due to redundancy? hand made CLD for comparisons

Thank you

score 2 · Accepted Answer · answered Nov 02 '20 at 01:35

Here's the example from survminer

library(survminer)
library(multcomp)
library(tidyr)
data(myeloma)

res <- pairwise_survdiff(Surv(time, event) ~ molecular_group,
                         data = myeloma)

Looking at the internals of the glht.summary method from the multcomp package, we create the lvl_order vector which identifies the ordering of the levels of x from smallest to largest.

x <- myeloma$molecular_group
levs <- levels(x)
y <- Surv(myeloma$time, myeloma$event)
lvl_order <- levels(x)[order(tapply(as.numeric(y)[1:length(x)], 
                                        x, mean))]

Then we can re-arrange the p-values from the res object into a matrix. mycomps is a matrix of the two sides of the paired comparisons. The signif vector is logical indicating whether differences are significant or not.

comps <- as_tibble(res$p.value, rownames="row") %>% 
  pivot_longer(-row, names_to="col", values_to="p") %>% 
  na.omit()
mycomps <- as.matrix(comps[,1:2])
signif <- comps$p < .05

Then, you can use the insert_absorb internal function to create the letters:

multcomp:::insert_absorb(signif, 
                         decreasing=FALSE, 
                         comps=mycomps, 
                         lvl_order=lvl_order)
# $Letters
# MAF    Proliferation       Cyclin D-2            MMSET     Hyperdiploid 
# "ab"              "a"              "b"             "ab"              "b" 
# Low bone disease       Cyclin D-1 
# "ab"             "ab" 
# 
# $monospacedLetters
# MAF    Proliferation       Cyclin D-2            MMSET     Hyperdiploid 
# "ab"             "a "             " b"             "ab"             " b" 
# Low bone disease       Cyclin D-1 
# "ab"             "ab" 
# 
# $LetterMatrix
# a     b
# MAF               TRUE  TRUE
# Proliferation     TRUE FALSE
# Cyclin D-2       FALSE  TRUE
# MMSET             TRUE  TRUE
# Hyperdiploid     FALSE  TRUE
# Low bone disease  TRUE  TRUE
# Cyclin D-1        TRUE  TRUE
# 
# $aLetters
# [1] "a" "b" "c" "d" "e" "f" "g" "h" "i" "j" "k" "l" "m" "n" "o" "p" "q" "r" "s" "t" "u" "v"
# [23] "w" "x" "y" "z" "A" "B" "C" "D" "E" "F" "G" "H" "I" "J" "K" "L" "M" "N" "O" "P" "Q" "R"
# [45] "S" "T" "U" "V" "W" "X" "Y" "Z"
# 
# $aseparator
# [1] "."
# 
# attr(,"class")
# [1] "multcompLetters"

Compact Letter Display from a matrix of significancies or by hand

1 Answers1